Whisper v3 for Enterprise Speech AI: Architecture & Integration

from faster_whisper import WhisperModel
import json

def transcribe_audio_whisper_v3(audio_file_path: str) -> dict:
  """
  Transcribes audio file using self-hosted Whisper Large-v3 via faster-whisper CTranslate2 runtime.
  """
  model_size = "large-v3"
  
  # Initialize model on GPU with INT8 compute precision
  model = WhisperModel(
      model_size, 
      device="cuda", 
      compute_type="int8",
      download_root="./whisper_models"
  )

  # Execute transcription with Silero VAD pre-filtering
  segments, info = model.transcribe(
      audio_file_path,
      beam_size=5,
      vad_filter=True,
      vad_parameters=dict(min_silence_duration_ms=500),
      word_timestamps=True
  )

  results = []
  for segment in segments:
      words_data = [
          {"word": w.word, "start": round(w.start, 2), "end": round(w.end, 2)}
          for w in segment.words
      ]
      results.append({
          "start": round(segment.start, 2),
          "end": round(segment.end, 2),
          "text": segment.text.strip(),
          "words": words_data
      })

  return {
      "language": info.language,
      "language_probability": round(info.language_probability, 4),
      "duration": round(info.duration, 2),
      "segments": results
  }

if __name__ == "__main__":
  audio_path = "./samples/customer_support_call.wav"
  output = transcribe_audio_whisper_v3(audio_path)
  print("Whisper v3 Output:", json.dumps(output, indent=2)[:400])